A deep convolutional neural network for vibration-based health-monitoring of rotating machinery

Abstract

The gearbox is a critical component in the mechanical system, requiring vigilant monitoring to prevent adverse consequences on safety and quality due to malfunction. Therefore, early fault diagnosis of the gearbox before the fatal breakdown of the entire mechanical system is of imperative importance. This study proposes a one-dimensional deep convolutional neural network (1D-DCNN) to learn features directly from the vibrational signals and identify the gear fault under different health conditions. The performance is compared with the decision tree, random forest, and support vector machine to validate the superiority of the 1D-DCNN. Experimental results showed that the proposed scheme outperforms other comparative methods, with a diagnostic accuracy of 97.11 %, thus confirming its effectiveness.